Netta1994/setfit_baai_cybereason_gpt-4o_improved-cot-instructions_two_reasoning_remove_final_ev

SetFit with BAAI/bge-base-en-v1.5

This is a SetFit model that can be used for Text Classification. This SetFit model uses BAAI/bge-base-en-v1.5 as the Sentence Transformer embedding model. A LogisticRegression instance is used for classification.

The model has been trained using an efficient few-shot learning technique that involves:

1. Fine-tuning a Sentence Transformer with contrastive learning.
2. Training a classification head with features from the fine-tuned Sentence Transformer.

Model Details

Model Description

• Model Type: SetFit
• Sentence Transformer body: BAAI/bge-base-en-v1.5
• Classification head: a LogisticRegression instance
• Maximum Sequence Length: 512 tokens
• Number of Classes: 2 classes

Model Sources

• Repository: SetFit on GitHub
• Paper: Efficient Few-Shot Learning Without Prompts
• Blogpost: SetFit: Efficient Few-Shot Learning Without Prompts

Model Labels

Label	Examples
0	'Reasoning why the answer may be good:\n- A correct definition of ESG including explaining its aspects (Environmental, Social, and Governance) is provided, which is discussed in the document.\n\nReasoning why the answer may be bad:\n- The answer does not address the specific question asked, which is about measures taken by Chuck Studios to promote sustainability and reduce waste. The document does not mention Chuck Studios at all, making the answer entirely irrelevant to the question.\n\nFinal Result:' 'Reasoning why the answer may be good:\n\n1. Context Grounding: The answer follows the main steps outlined in the document, such as cutting the carrots, adding water, covering with a lid or plastic wrap, microwaving, and checking for tenderness.\n2. Relevance: It addresses the question directly by providing a method to microwave carrots.\n3. Conciseness: The answer is straightforward and does not deviate into unrelated topics.\n\nReasoning why the answer may be bad:\n\n1. Context Grounding: The document does not mention anything about carrots changing color to blue when microwaved for too long; this information is fabricated and not supported by the source.\n2. Conciseness: While mostly concise, the inclusion of the incorrect information about carrots turning blue adds unnecessary and misleading information.\n\nFinal Result: ****' "Reasoning why the answer may be good:\n1. Context Grounding: The answer is well-grounded in the context provided by the document. It accurately reflects the steps for enabling approval for appointment bookings as described in the provided text.\n2. Relevance: The answer addresses the method of enabling appointment approval, indirectly solving the problem where clients might be unable to book their appointments online.\n3. Conciseness: While the answer is detailed, it is somewhat concise in outlining the step-by-step process.\n\nReasoning why the answer may be bad:\n1. Context Grounding: The answer provided is not directly addressing the specific issue of why clients are unable to book appointments online. Instead, it describes how to enable appointment approval.\n2. Relevance: The answer diverts from the question asked. It does not diagnose or suggest troubleshooting steps for the issue of clients being unable to book appointments.\n3. Conciseness: The answer includes a step-by-step process that isn't required unless it directly solves the inability to book problem. It adds unnecessary steps without first determining if those stepsare relevant to the issue.\n\nFinal result: ****"
1	'**Reasoning:\n\nWhy the answer may be good:\n1. Context Grounding: The answer is well-supported by the document, which lists the specific benefits the author has experienced from their yoga practice.\n2. Relevance: The answer directly addresses the question by listing the benefits from the yoga practice, which is exactly what was asked.\n3. Conciseness: The answer is to the point and only includes the relevant benefits without unnecessary information.\n\nWhy the answer may be bad:**\n1. Context Grounding: There is no issue here; the answer is consistent with the document.\n2. Relevance: There is no deviation from the topic; each listed benefit is directly taken from the document.\n3. Conciseness: The answer mentions all 10 points clearly and without excessive elaboration, maintaining conciseness.\n\nFinal Result: ****' 'Reasoning:\nThe given answer is relatively good. \n\n1. Context Grounding: The answer accurately captures several points highlighted in the document. It addresses key characteristics such as fear of gaining weight, food refusal, obsession with food, making excuses not to eat, mood swings, and signs of depression or anxiety.\n2. Relevance: The answer is relevant and directly answers the question about identifying signs of anorexia.\n3. Conciseness: The answer is concise and avoids delving into peripheral details, focusing instead on providing actionable signs to look out for.\n\nHowever, the answer could have included more signs like unusual facial or body hair, sensitivity to cold, or increased exercise habits to be comprehensive.\n\nFinal Result:' 'Reasoning:\n1. Context Grounding: The answer is closely derived from the provided document, where it lists steps to address the "submitted url marked noindex" error message on a site inspection page.\n2. Relevance: The answer partially addresses the specific question asked regarding what to do if all pages are excluded. It focuses mainly on resolving the 'noindex' issue rather than handling a situation where all pages are excluded.\n3. Conciseness: The answer is clear and to the point, though it is detailed enough to cover the method of resolving the 'noindex' problem in a structured manner. However, it may not fully address the broader query of exclusion of all pages.\n4. Correct and Detailed Instructions: The steps included are detailed and correct for resolving a 'noindex' issue, but they may not fully cover other potential causes for all pages being excluded.\n\nFinal Result:'

Evaluation

Metrics

Label	Accuracy
all	0.7042

Uses

Direct Use for Inference

First install the SetFit library:

pip install setfit

Then you can load this model and run inference.

from setfit import SetFitModel

# Download from the 🤗 Hub
model = SetFitModel.from_pretrained("Netta1994/setfit_baai_cybereason_gpt-4o_improved-cot-instructions_two_reasoning_remove_final_ev")
# Run inference
preds = model("Reasoning why the answer may be good:
- The answer provides a specific URL, which is required by the question.
- It appears to be in the format expected for image URLs as hinted at in the document.

Reasoning why the answer may be bad:
- The provided answer does not match the precise URL given in the document.
- The correct URL for the second query should be `..\/..\/_images\/hunting_http://miller.co`, while the answer contains `hunting_http://www.flores.net/`, which is not mentioned in the document.
- The answer does not reflect careful cross-referencing with the provided document.

Final result:")

Training Details

Training Set Metrics

Training set	Min	Median	Max
Word count	45	136.1487	302

Label	Training Sample Count
0	311
1	321

Training Hyperparameters

• batch_size: (16, 16)
• num_epochs: (1, 1)
• max_steps: -1
• sampling_strategy: oversampling
• num_iterations: 20
• body_learning_rate: (2e-05, 2e-05)
• head_learning_rate: 2e-05
• loss: CosineSimilarityLoss
• distance_metric: cosine_distance
• margin: 0.25
• end_to_end: False
• use_amp: False
• warmup_proportion: 0.1
• l2_weight: 0.01
• seed: 42
• eval_max_steps: -1
• load_best_model_at_end: False

Training Results

Epoch	Step	Training Loss	Validation Loss
0.0006	1	0.1893	-
0.0316	50	0.2645	-
0.0633	100	0.2581	-
0.0949	150	0.2491	-
0.1266	200	0.2544	-
0.1582	250	0.2538	-
0.1899	300	0.2413	-
0.2215	350	0.1942	-
0.2532	400	0.1354	-
0.2848	450	0.0857	-
0.3165	500	0.0544	-
0.3481	550	0.0412	-
0.3797	600	0.0313	-
0.4114	650	0.0239	-
0.4430	700	0.018	-
0.4747	750	0.0268	-
0.5063	800	0.0185	-
0.5380	850	0.0245	-
0.5696	900	0.0255	-
0.6013	950	0.0201	-
0.6329	1000	0.0187	-
0.6646	1050	0.0132	-
0.6962	1100	0.0129	-
0.7278	1150	0.0065	-
0.7595	1200	0.004	-
0.7911	1250	0.0029	-
0.8228	1300	0.0028	-
0.8544	1350	0.0026	-
0.8861	1400	0.0022	-
0.9177	1450	0.0021	-
0.9494	1500	0.0021	-
0.9810	1550	0.0019	-

Framework Versions

• Python: 3.10.14
• SetFit: 1.1.0
• Sentence Transformers: 3.1.1
• Transformers: 4.44.0
• PyTorch: 2.4.0+cu121
• Datasets: 3.0.0
• Tokenizers: 0.19.1

Citation

BibTeX

@article{https://doi.org/10.48550/arxiv.2209.11055,
    doi = {10.48550/ARXIV.2209.11055},
    url = {https://arxiv.org/abs/2209.11055},
    author = {Tunstall, Lewis and Reimers, Nils and Jo, Unso Eun Seo and Bates, Luke and Korat, Daniel and Wasserblat, Moshe and Pereg, Oren},
    keywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, FOS: Computer and information sciences},
    title = {Efficient Few-Shot Learning Without Prompts},
    publisher = {arXiv},
    year = {2022},
    copyright = {Creative Commons Attribution 4.0 International}
}